Development and validation of a new algorithm for attribution of neuropsychiatric events in systemic lupus erythematosus
Development and validation of a new algorithm for attribution of neuropsychiatric events in systemic lupus erythematosus
复制标题
DOI：
10.1093/rheumatology/keu384
复制
发表时间：
2015-05-01
期刊：
影响因子：
5.5
通讯作者：
Govoni, Marcello
中科院分区：
文献类型：
--
作者：
Bortoluzzi, Alessandra;Scire, Carlo Alberto;Govoni, Marcello
目标。本研究的目的是开发和验证一种算法，以帮助将SLE患者的神经精神（NP）事件归因于潜在疾病。第一阶段通过文献信息一致驱动的过程，确定并分类候选项目，将其纳入将NP事件归因为SLE的算法及其相对权重。第二阶段采用系统性红斑狼疮的回顾性培训队列，通过数据驱动的过程验证第一阶段选择的项目，并细化权重，拟合项目作为自变量，专家评价（临床判断）作为逻辑模型的参考标准。第3期包括使用外部多中心回顾性SLE队列的验证过程。第一阶段确定了四个不同的项目：NP事件的时间、事件类型、混杂因素和有利因素。培训队列和验证队列分别包括228和221名患者。每个患者至少经历了一次以ACR病例定义为特征的NP事件。在这些样本中，第一阶段选择的项目在区分NPSLE患者方面表现良好：使用二分类结果的受试者工作特征曲线下面积在训练集中为0.87，在验证集中为0.82。在训练组和验证组中，验证评分的相关截断值分别确定阳性预测值为100% （95% CI 93.2, 100）和86.3% （95% CI 76.2, 93.2）的事件。研究人员开发并验证了一种基于概率评分的新算法，以确定NP事件与SLE之间的关系。
Objective. The aim of this study was to develop and validate an algorithm to assist the attribution of neuropsychiatric (NP) events to underlying disease in SLE patients.Methods. Phase 1 identified and categorized candidate items to be included in the algorithm for the attribution of an NP event to SLE and their relative weights through a literature-informed consensus-driven process. Using a retrospective training cohort of SLE, phase 2 validated items selected in phase 1 and refined weights through a data-driven process, fitting items as independent variables and expert evaluation (clinical judgement) as reference standard in logistic models. Phase 3 consisted of a validation process using an external multicentre retrospective SLE cohort.Results. Phase 1 identified four different items: timing of the NP event, type of event, confounding factors and favouring factors. The training and validating cohorts included 228 and 221 patients, respectively. Each patient experienced at least one NP event characterized using the ACR case definition. In these samples, items selected in phase 1 showed good performance in discriminating patients with NPSLE: the area under the receiver operating characteristic curve using dichotomous outcomes was 0.87 in the training set and 0.82 in the validating set. Relevant cut-offs of the validated score identify events with a positive predictive value of 100% (95% CI 93.2, 100) and 86.3% (95% CI 76.2, 93.2) in the training and validating cohorts, respectively.Conclusion. A new algorithm based on a probability score was developed and validated to determine the relationship between NP events and SLE.